What we got right
Support at real scale, in every language, around the clock.
We connected their backend systems and handled support across 30 countries. On paper, it was a major technical success.
€50M+
revenue
30
countries
4,000+
messages a day
Our blueprint
What losing a customer
taught us.
We helped a €50M+ multi-country retailer automate thousands of daily messages, and then we lost them. Here is exactly why, and how that failure became the blueprint for everything we do now.
Why we lost them · 1
The bill grew every time they succeeded.
We charged per ticket. The more we helped, the more it cost. The value was invisible; the cost was on every statement. When budgets tightened, we were the easy line to cut.
JANINVOICEAUG
Why we lost them · 2
It became a black box.
Nobody was truly watching the AI. When something went wrong, they heard it from their customers, not from us. No supervisor, no anomaly watcher, no morning briefing.
The lesson
A good tool was never going to be enough.
The technology worked. What was missing was everything around it. Solutions like this have to be built into a company, its systems, its processes and its people. That takes three kinds of expertise.
Expertise in what you cover
Order tracking across carriers, returns and reklamacije, payment mix-ups, peak season. And which questions should never get an automatic answer.
Expertise in your company
The answers live in your shop, ERP and warehouse, and in rules only your team knows. All of it has to be mapped and connected, not installed.
Expertise in change
AI in a live support operation is a change project: step by step, tested before customers see it, measured and improved every week.
We sell the system we wish we'd built the first time.
Then → now
Per-ticket bills that punished growth
A capped monthly price. You know your maximum before you sign.
A black box nobody could see into
A supervisor on every answer. Cross-checked against your data before it reaches a customer.
Problems surfaced by their customers
An anomaly watcher. Flags refund and fraud patterns and silent failures, and freezes risky actions.
No idea what the system was doing
A daily briefing. What happened, what we caught, what needs you. Ask it anything.
Deploy-and-walk-away software
Our team in Slovenia. Runs it, tunes it and stands behind it every day.
Live right now · rollout in progress
A leading Slovenian online retailer, running across two markets.
Real numbers from July and August 2026.
7,600+
of 7,850+ conversations answered by our AI team, on web chat and email
€7,000+
of support capacity over the two months, counted conservatively
600+ h
of agent time given back, about two full-time agents every month
Seconds
typical first reply, day or night, and under a minute on average in August
2,700
answered outside business hours, nights and weekends included, in 6 languages
9
unusual patterns caught by the anomaly watcher, each flagged the moment it started
How we work with them, week to week
01
Step by step, across two markets
One twin, web chat and email. On email we started with the messages we handle well; the rest still goes to their ticket system, untouched.
02
Knowledge fixed at the source
When an answer is wrong, we fix where it came from: the document, the rule or the policy, not the single reply.
03
A knowledge harvester, every night
It sorts every gap the AI couldn't handle. Their knowledge owner only gets questions they can actually answer.
04
Every handover reviewed
Should it have been answered, can we fix it, is it a new gap, or was a person the right call?
05
Tested and measured, not assumed
Dozens of real questions re-asked every night and after every change, compared against a market where nothing changed.
06
In control, every day
Live Control Room and morning briefing. Legal threats and chargebacks go straight to a person.